Origins & Computability · 2004
A Tutorial Introduction to the Minimum Description Length Principle
Gave the standard accessible account of the minimum description length principle, learning as finding the hypothesis that compresses the data best, and explained how refined codes based on normalised maximum likelihood replace crude two-part codes, linking Kolmogorov complexity to practical model selection.
Editorial record
Plain-language summary
MDL says that a good model is one that lets you describe your data in the fewest bits, counting the bits needed to describe the model itself. Grünwald's tutorial walks from that intuition to the mathematics: two-part codes, universal codes, and the normalised maximum likelihood distribution that makes the idea precise. It is the usual entry point for readers who want the formal basis of the claim that compression and prediction are two sides of the same coin, which is why it sits on reading lists next to Hinton's MDL paper and Kolmogorov complexity.
Source record
Provenance
- Record ID
- P-682
- Record created
- 2026-09-26
- Last reviewed
- 2026-09-26
- Record version
- 1
- https://arxiv.org/abs/math/0406077
- arXiv:math/0406077
Citation caveat: Citation metadata is approximate and marked unverified in the source dataset.